This patch corresponds to review:
http://reviews.llvm.org/D9440
It adds a new register class to the PPC back end to contain single precision
values in VSX registers. Additionally, it adds scalar loads and stores for
VSX registers.
llvm-svn: 236755
This patch adds Hardware Transaction Memory (HTM) support supported by ISA 2.07
(POWER8). The intrinsic support is based on GCC one [1], but currently only the
'PowerPC HTM Low Level Built-in Function' are implemented.
The HTM instructions follows the RC ones and the transaction initiation result
is set on RC0 (with exception of tcheck). Currently approach is to create a
register copy from CR0 to GPR and comapring. Although this is suboptimal, since
the branch could be taken directly by comparing the CR0 value, it generates code
correctly on both test and branch and just return value. A possible future
optimization could be elimitate the MFCR instruction to branch directly.
The HTM usage requires a recently newer kernel with PPC HTM enabled. Tested on
powerpc64 and powerpc64le.
This is send along a clang patch to enabled the builtins and option switch.
[1] https://gcc.gnu.org/onlinedocs/gcc/PowerPC-Hardware-Transactional-Memory-Built-in-Functions.html
Phabricator Review: http://reviews.llvm.org/D8247
llvm-svn: 233204
This adds support for the QPX vector instruction set, which is used by the
enhanced A2 cores on the IBM BG/Q supercomputers. QPX vectors are 256 bytes
wide, holding 4 double-precision floating-point values. Boolean values, modeled
here as <4 x i1> are actually also represented as floating-point values
(essentially { -1, 1 } for { false, true }). QPX shares many features with
Altivec and VSX, but is distinct from both of them. One major difference is
that, instead of adding completely-separate vector registers, QPX vector
registers are extensions of the scalar floating-point registers (lane 0 is the
corresponding scalar floating-point value). The operations supported on QPX
vectors mirrors that supported on the scalar floating-point values (with some
additional ones for permutations and logical/comparison operations).
I've been maintaining this support out-of-tree, as part of the bgclang project,
for several years. This is not the entire bgclang patch set, but is most of the
subset that can be cleanly integrated into LLVM proper at this time. Adding
this to the LLVM backend is part of my efforts to rebase bgclang to the current
LLVM trunk, but is independently useful (especially for codes that use LLVM as
a JIT in library form).
The assembler/disassembler test coverage is complete. The CodeGen test coverage
is not, but I've included some tests, and more will be added as follow-up work.
llvm-svn: 230413
Our register allocation has become better recently, it seems, and is now
starting to generate cross-block copies into inflated register classes. These
copies are not transformed into subregister insertions/extractions by the
PPCVSXCopy class, and so need to be handled directly by
PPCInstrInfo::copyPhysReg. The code to do this was *almost* there, but not
quite (it was unnecessarily restricting itself to only the direct
sub/super-register-class case (not copying between, for example, something in
VRRC and the lower-half of VSRC which are super-registers of F8RC).
Triggering this behavior manually is difficult; I'm including two
bugpoint-reduced test cases from the test suite.
llvm-svn: 229457
PPCInstrInfo.cpp has ended up containing several small MI-level passes, and
this is making the file harder to read than necessary. Split out
PPCEarlyReturn into its own source file. NFC.
Now that PPCInstrInfo.cpp does not also contain pass implementations, I hope
that it will be slightly less unwieldy.
llvm-svn: 227775
PPCInstrInfo.cpp has ended up containing several small MI-level passes, and
this is making the file harder to read than necessary. Split out
PPCVSXCopy into its own source file. NFC.
llvm-svn: 227771
PPCInstrInfo.cpp has ended up containing several small MI-level passes, and
this is making the file harder to read than necessary. Split out
PPCVSXFMAMutate into its own source file. NFC.
llvm-svn: 227770
This MI-level pass was necessary when VSX support was first being developed,
specifically, before the ABI code had been updated to use VSX registers for
arguments (the register assignments did not change, in a physical sense, but
the VSX super-registers are now used). Unfortunately, I never went back and
removed this pass after that was done. I believe this code is now effectively
dead.
llvm-svn: 227767
When PPCEarlyReturn, it should really copy implicit ops from the old return
instruction to the new one. This currently does not matter much, because we run
PPCEarlyReturn very late in the pipeline (there is nothing to do DCE on
definitions of those registers). However, for completeness, we should do it
anyway.
Noticed by inspection (and there should be no functional change); thus, no
test case.
llvm-svn: 227763
This re-applies r225808, fixed to avoid problems with SDAG dependencies along
with the preceding fix to ScheduleDAGSDNodes::RegDefIter::InitNodeNumDefs.
These problems caused the original regression tests to assert/segfault on many
(but not all) systems.
Original commit message:
This commit does two things:
1. Refactors PPCFastISel to use more of the common infrastructure for call
lowering (this lets us take advantage of this common code for lowering some
common intrinsics, stackmap/patchpoint among them).
2. Adds support for stackmap/patchpoint lowering. For the most part, this is
very similar to the support in the AArch64 target, with the obvious differences
(different registers, NOP instructions, etc.). The test cases are adapted
from the AArch64 test cases.
One difference of note is that the patchpoint call sequence takes 24 bytes, so
you can't use less than that (on AArch64 you can go down to 16). Also, as noted
in the docs, we take the patchpoint address to be the actual code address
(assuming the call is local in the TOC-sharing sense), which should yield
higher performance than generating the full cross-DSO indirect-call sequence
and is likely just as useful for JITed code (if not, we'll change it).
StackMaps and Patchpoints are still marked as experimental, and so this support
is doubly experimental. So go ahead and experiment!
llvm-svn: 225909
This commit does two things:
1. Refactors PPCFastISel to use more of the common infrastructure for call
lowering (this lets us take advantage of this common code for lowering some
common intrinsics, stackmap/patchpoint among them).
2. Adds support for stackmap/patchpoint lowering. For the most part, this is
very similar to the support in the AArch64 target, with the obvious differences
(different registers, NOP instructions, etc.). The test cases are adapted
from the AArch64 test cases.
One difference of note is that the patchpoint call sequence takes 24 bytes, so
you can't use less than that (on AArch64 you can go down to 16). Also, as noted
in the docs, we take the patchpoint address to be the actual code address
(assuming the call is local in the TOC-sharing sense), which should yield
higher performance than generating the full cross-DSO indirect-call sequence
and is likely just as useful for JITed code (if not, we'll change it).
StackMaps and Patchpoints are still marked as experimental, and so this support
is doubly experimental. So go ahead and experiment!
llvm-svn: 225808
We really need a separate 64-bit version of this instruction so that it can be
marked as clobbering LR8 (instead of just LR). No change in functionality
(although the verifier might be slightly happier), however, it is required for
stackmap/patchpoint support. Thus, this will be covered by stackmap test cases
once those are added.
llvm-svn: 225804
On PPC64, we end up with lots of i32 -> i64 zero extensions, not only from all
of the usual places, but also from the ABI, which specifies that values passed
are zero extended. Almost all 32-bit PPC instructions in PPC64 mode are defined
to do *something* to the higher-order bits, and for some instructions, that
action clears those bits (thus providing a zero-extended result). This is
especially common after rotate-and-mask instructions. Adding an additional
instruction to zero-extend the results of these instructions is unnecessary.
This PPCISelDAGToDAG peephole optimization examines these zero-extensions, and
looks back through their operands to see if all instructions will implicitly
zero extend their results. If so, we convert these instructions to their 64-bit
variants (which is an internal change only, the actual encoding of these
instructions is the same as the original 32-bit ones) and remove the
unnecessary zero-extension (changing where the INSERT_SUBREG instructions are
to make everything internally consistent).
llvm-svn: 224169
With VSX enabled, test/CodeGen/PowerPC/recipest.ll exposes a bug in
the FMA mutation pass. If we have a situation where a killed product
register is the same register as the FMA target, such as:
%vreg5<def,tied1> = XSNMSUBADP %vreg5<tied0>, %vreg11, %vreg5,
%RM<imp-use>; VSFRC:%vreg5 F8RC:%vreg11
then the substitution makes no sense. We end up getting a crash when
we try to extend the interval associated with the killed product
register, as there is already a live range for %vreg5 there. This
patch just disables the mutation under those circumstances.
Since recipest.ll generates different code with VMX enabled, I've
modified that test to use -mattr=-vsx. I've borrowed the code from
that test that exposed the bug and placed it in fma-mutate.ll, where
it tests several mutation opportunities including the "bad" one.
llvm-svn: 220290
With VSX enabled, LLVM crashes when compiling
test/CodeGen/PowerPC/fma.ll. I traced this to the liveness test
that's revised in this patch. The interval test is designed to only
work for virtual registers, but in this case the AddendSrcReg is
physical. Since there is already a walk of the MIs between the
AddendMI and the FMA, I added a check for def/kill of the AddendSrcReg
in that loop. At Hal Finkel's request, I converted the liveness test
to an assert restricted to virtual registers.
I've changed the fma.ll test to have VSX and non-VSX variants so we
can test both kinds of multiply-adds.
llvm-svn: 220090
includes handling DIR_PWR8 where appropriate
The P7Model Itinerary is currently tied in for use under the P8Model, and will be updated later.
llvm-svn: 211779
system headers above the includes of generated '.inc' files that
actually contain code. In a few targets this was already done pretty
consistently, but it wasn't done *really* consistently anywhere. It is
strictly cleaner IMO and necessary in a bunch of places where the
DEBUG_TYPE is referenced from the generated code. Consistency with the
necessary places trumps. Hopefully the build bots are OK with the
movement of intrin.h...
llvm-svn: 206838
behavior based on other files defining DEBUG_TYPE, which means it cannot
define DEBUG_TYPE at all. This is actually better IMO as it forces folks
to define relevant DEBUG_TYPEs for their files. However, it requires all
files that currently use DEBUG(...) to define a DEBUG_TYPE if they don't
already. I've updated all such files in LLVM and will do the same for
other upstream projects.
This still leaves one important change in how LLVM uses the DEBUG_TYPE
macro going forward: we need to only define the macro *after* header
files have been #include-ed. Previously, this wasn't possible because
Debug.h required the macro to be pre-defined. This commit removes that.
By defining DEBUG_TYPE after the includes two things are fixed:
- Header files that need to provide a DEBUG_TYPE for some inline code
can do so by defining the macro before their inline code and undef-ing
it afterward so the macro does not escape.
- We no longer have rampant ODR violations due to including headers with
different DEBUG_TYPE definitions. This may be mostly an academic
violation today, but with modules these types of violations are easy
to check for and potentially very relevant.
Where necessary to suppor headers with DEBUG_TYPE, I have moved the
definitions below the includes in this commit. I plan to move the rest
of the DEBUG_TYPE macros in LLVM in subsequent commits; this one is big
enough.
The comments in Debug.h, which were hilariously out of date already,
have been updated to reflect the recommended practice going forward.
llvm-svn: 206822
We had stored both f64 values and v2f64, etc. values in the VSX registers. This
worked, but was suboptimal because we would always spill 16-byte values even
through we almost always had scalar 8-byte values. This resulted in an
increase in stack-size use, extra memory bandwidth, etc. To fix this, I've
added 64-bit subregisters of the Altivec registers, and combined those with the
existing scalar floating-point registers to form a class of VSX scalar
floating-point registers. The ABI code has also been enhanced to use this
register class and some other necessary improvements have been made.
llvm-svn: 205075
As explained in r204976, because of how the allocation of VSX registers
interacts with the call-lowering code, we sometimes end up generating self VSX
copies. Specifically, things like this:
%VSL2<def> = COPY %F2, %VSL2<imp-use,kill>
(where %F2 is really a sub-register of %VSL2, and so this copy is a nop)
This adds a small cleanup pass to remove these prior to post-RA scheduling.
llvm-svn: 204980
Because of how the allocation of VSX registers interacts with the call-lowering
code, we sometimes end up generating self VSX copies. Specifically, things like
this:
%VSL2<def> = COPY %F2, %VSL2<imp-use,kill>
(where %F2 is really a sub-register of %VSL2, and so this copy is a nop)
The problem is that ExpandPostRAPseudos always assumes that *some* instruction
has been inserted, and adds implicit defs to it. This is a problem if no copy
was inserted because it can cause subtle problems during post-RA scheduling.
These self copies will have to be removed some other way.
llvm-svn: 204976
The VSX instruction set has two types of FMA instructions: A-type (where the
addend is taken from the output register) and M-type (where one of the product
operands is taken from the output register). This adds a small pass that runs
just after MI scheduling (and, thus, just before register allocation) that
mutates A-type instructions (that are created during isel) into M-type
instructions when:
1. This will eliminate an otherwise-necessary copy of the addend
2. One of the product operands is killed by the instruction
The "right" moment to make this decision is in between scheduling and register
allocation, because only there do we know whether or not one of the product
operands is killed by any particular instruction. Unfortunately, this also
makes the implementation somewhat complicated, because the MIs are not in SSA
form and we need to preserve the LiveIntervals analysis.
As a simple example, if we have:
%vreg5<def> = COPY %vreg9; VSLRC:%vreg5,%vreg9
%vreg5<def,tied1> = XSMADDADP %vreg5<tied0>, %vreg17, %vreg16,
%RM<imp-use>; VSLRC:%vreg5,%vreg17,%vreg16
...
%vreg9<def,tied1> = XSMADDADP %vreg9<tied0>, %vreg17, %vreg19,
%RM<imp-use>; VSLRC:%vreg9,%vreg17,%vreg19
...
We can eliminate the copy by changing from the A-type to the
M-type instruction. This means:
%vreg5<def,tied1> = XSMADDADP %vreg5<tied0>, %vreg17, %vreg16,
%RM<imp-use>; VSLRC:%vreg5,%vreg17,%vreg16
is replaced by:
%vreg16<def,tied1> = XSMADDMDP %vreg16<tied0>, %vreg18, %vreg9,
%RM<imp-use>; VSLRC:%vreg16,%vreg18,%vreg9
and we remove: %vreg5<def> = COPY %vreg9; VSLRC:%vreg5,%vreg9
llvm-svn: 204768
Although the first two operands are the ones that can be swapped, the tied
input operand is listed before them, so we need to adjust for that.
I have a test case for this, but it goes along with an upcoming commit (so it
will come soon).
llvm-svn: 204748
I've done some experimentation with this, and it looks like using the
lower-latency (but lower throughput) copy instruction is essentially always the
right thing to do.
My assumption is that, in order to be relatively sure that the higher-latency
copy will increase throughput, we'd want to have it unlikely to be in-flight
with its use. On the P7, the global completion table (GCT) can hold a maximum
of 120 instructions, shared among all active threads (up to 4), giving 30
instructions per thread. So specifically, I'd require at least that many
instructions between the copy and the use before the high-latency variant is
used.
Trying this, however, over the entire test suite resulted in zero cases where
the high-latency form would be preferable. This may be a consequence of the
fact that the scheduler views copies as free, and so they tend to end up close
to their uses. For this experiment I created a function:
unsigned chooseVSXCopy(MachineBasicBlock &MBB,
MachineBasicBlock::iterator I,
unsigned DestReg, unsigned SrcReg,
unsigned StartDist = 1,
unsigned Depth = 3) const;
with an implementation like:
if (!Depth)
return PPC::XXLOR;
const unsigned MaxDist = 30;
unsigned Dist = StartDist;
for (auto J = I, JE = MBB.end(); J != JE && Dist <= MaxDist; ++J) {
if (J->isTransient() && !J->isCopy())
continue;
if (J->isCall() || J->isReturn() || J->readsRegister(DestReg, TRI))
return PPC::XXLOR;
++Dist;
}
// We've exceeded the required distance for the high-latency form, use it.
if (Dist > MaxDist)
return PPC::XVCPSGNDP;
// If this is only an exit block, use the low-latency form.
if (MBB.succ_empty())
return PPC::XXLOR;
// We've reached the end of the block, check the successor blocks (up to some
// depth), and use the high-latency form if that is okay with all successors.
for (auto J = MBB.succ_begin(), JE = MBB.succ_end(); J != JE; ++J) {
if (chooseVSXCopy(**J, (*J)->begin(), DestReg, SrcReg,
Dist, --Depth) == PPC::XXLOR)
return PPC::XXLOR;
}
// All of our successor blocks seem okay with the high-latency variant, so
// we'll use it.
return PPC::XVCPSGNDP;
and then changed the copy opcode selection from:
Opc = PPC::XXLOR;
to:
Opc = chooseVSXCopy(MBB, std::next(I), DestReg, SrcReg);
In conclusion, I'm removing the FIXME from the comment, because I believe that
there is, at least absent other examples, nothing to fix.
llvm-svn: 204591
operator* on the by-operand iterators to return a MachineOperand& rather than
a MachineInstr&. At this point they almost behave like normal iterators!
Again, this requires making some existing loops more verbose, but should pave
the way for the big range-based for-loop cleanups in the future.
llvm-svn: 203865
VSX is an ISA extension supported on the POWER7 and later cores that enhances
floating-point vector and scalar capabilities. Among other things, this adds
<2 x double> support and generally helps to reduce register pressure.
The interesting part of this ISA feature is the register configuration: there
are 64 new 128-bit vector registers, the 32 of which are super-registers of the
existing 32 scalar floating-point registers, and the second 32 of which overlap
with the 32 Altivec vector registers. This makes things like vector insertion
and extraction tricky: this can be free but only if we force a restriction to
the right register subclass when needed. A new "minipass" PPCVSXCopy takes care
of this (although it could do a more-optimal job of it; see the comment about
unnecessary copies below).
Please note that, currently, VSX is not enabled by default when targeting
anything because it is not yet ready for that. The assembler and disassembler
are fully implemented and tested. However:
- CodeGen support causes miscompiles; test-suite runtime failures:
MultiSource/Benchmarks/FreeBench/distray/distray
MultiSource/Benchmarks/McCat/08-main/main
MultiSource/Benchmarks/Olden/voronoi/voronoi
MultiSource/Benchmarks/mafft/pairlocalalign
MultiSource/Benchmarks/tramp3d-v4/tramp3d-v4
SingleSource/Benchmarks/CoyoteBench/almabench
SingleSource/Benchmarks/Misc/matmul_f64_4x4
- The lowering currently falls back to using Altivec instructions far more
than it should. Worse, there are some things that are scalarized through the
stack that shouldn't be.
- A lot of unnecessary copies make it past the optimizers, and this needs to
be fixed.
- Many more regression tests are needed.
Normally, I'd fix these things prior to committing, but there are some
students and other contributors who would like to work this, and so it makes
sense to move this development process upstream where it can be subject to the
regular code-review procedures.
llvm-svn: 203768
This change enables tracking i1 values in the PowerPC backend using the
condition register bits. These bits can be treated on PowerPC as separate
registers; individual bit operations (and, or, xor, etc.) are supported.
Tracking booleans in CR bits has several advantages:
- Reduction in register pressure (because we no longer need GPRs to store
boolean values).
- Logical operations on booleans can be handled more efficiently; we used to
have to move all results from comparisons into GPRs, perform promoted
logical operations in GPRs, and then move the result back into condition
register bits to be used by conditional branches. This can be very
inefficient, because the throughput of these CR <-> GPR moves have high
latency and low throughput (especially when other associated instructions
are accounted for).
- On the POWER7 and similar cores, we can increase total throughput by using
the CR bits. CR bit operations have a dedicated functional unit.
Most of this is more-or-less mechanical: Adjustments were needed in the
calling-convention code, support was added for spilling/restoring individual
condition-register bits, and conditional branch instruction definitions taking
specific CR bits were added (plus patterns and code for generating bit-level
operations).
This is enabled by default when running at -O2 and higher. For -O0 and -O1,
where the ability to debug is more important, this feature is disabled by
default. Individual CR bits do not have assigned DWARF register numbers,
and storing values in CR bits makes them invisible to the debugger.
It is critical, however, that we don't move i1 values that have been promoted
to larger values (such as those passed as function arguments) into bit
registers only to quickly turn around and move the values back into GPRs (such
as happens when values are returned by functions). A pair of target-specific
DAG combines are added to remove the trunc/extends in:
trunc(binary-ops(binary-ops(zext(x), zext(y)), ...)
and:
zext(binary-ops(binary-ops(trunc(x), trunc(y)), ...)
In short, we only want to use CR bits where some of the i1 values come from
comparisons or are used by conditional branches or selects. To put it another
way, if we can do the entire i1 computation in GPRs, then we probably should
(on the POWER7, the GPR-operation throughput is higher, and for all cores, the
CR <-> GPR moves are expensive).
POWER7 test-suite performance results (from 10 runs in each configuration):
SingleSource/Benchmarks/Misc/mandel-2: 35% speedup
MultiSource/Benchmarks/Prolangs-C++/city/city: 21% speedup
MultiSource/Benchmarks/MiBench/automotive-susan: 23% speedup
SingleSource/Benchmarks/CoyoteBench/huffbench: 13% speedup
SingleSource/Benchmarks/Misc-C++/Large/sphereflake: 13% speedup
SingleSource/Benchmarks/Misc-C++/mandel-text: 10% speedup
SingleSource/Benchmarks/Misc-C++-EH/spirit: 10% slowdown
MultiSource/Applications/lemon/lemon: 8% slowdown
llvm-svn: 202451
As part of the cleanup done to enable the disassembler, the PPC instructions
now have a valid Size description field. This can now be used to replace some
custom logic in a few places to compute instruction sizes.
Patch by David Wiberg!
llvm-svn: 200623
GPRC_NOR0 is not a subclass of GPRC (because it also contains the ZERO pseudo
register). As a result, we also need to check for it in the spilling code.
llvm-svn: 200288
subsequent changes are easier to review. About to fix some layering
issues, and wanted to separate out the necessary churn.
Also comment and sink the include of "Windows.h" in three .inc files to
match the usage in Memory.inc.
llvm-svn: 198685
Without this, MachineCSE is powerless to handle redundant operations with truncated source operands.
This required fixing the 2-addr pass to handle tied subregisters. It isn't clear what combinations of subregisters can legally be tied, but the simple case of truncated source operands is now safely handled:
%vreg11<def> = COPY %vreg1:sub_32bit; GR32:%vreg11 GR64:%vreg1
%vreg12<def> = COPY %vreg2:sub_32bit; GR32:%vreg12 GR64:%vreg2
%vreg13<def,tied1> = ADD32rr %vreg11<tied0>, %vreg12<kill>, %EFLAGS<imp-def>
Test case: cse-add-with-overflow.ll.
This exposed an existing bug in
PPCInstrInfo::commuteInstruction. Thanks to Rafael for the test case:
PowerPC/crash.ll.
llvm-svn: 197465
Aside from a few minor latency corrections, the major change here is a new
hazard recognizer which focuses on better dispatch-group formation on the
POWER7. As with the PPC970's hazard recognizer, the most important thing it
does is avoid load-after-store hazards within the same dispatch group. It uses
the POWER7's special dispatch-group-terminating nop instruction (instead of
inserting multiple regular nop instructions). This new hazard recognizer makes
use of the scheduling dependency graph itself, built using AA information, to
robustly detect the possibility of load-after-store hazards.
significant test-suite performance changes (the error bars are 99.5% confidence
intervals based on 5 test-suite runs both with and without the change --
speedups are negative):
speedups:
MultiSource/Benchmarks/FreeBench/pcompress2/pcompress2
-0.55171% +/- 0.333168%
MultiSource/Benchmarks/TSVC/CrossingThresholds-dbl/CrossingThresholds-dbl
-17.5576% +/- 14.598%
MultiSource/Benchmarks/TSVC/Reductions-dbl/Reductions-dbl
-29.5708% +/- 7.09058%
MultiSource/Benchmarks/TSVC/Reductions-flt/Reductions-flt
-34.9471% +/- 11.4391%
SingleSource/Benchmarks/BenchmarkGame/puzzle
-25.1347% +/- 11.0104%
SingleSource/Benchmarks/Misc/flops-8
-17.7297% +/- 9.79061%
SingleSource/Benchmarks/Shootout-C++/ary3
-35.5018% +/- 23.9458%
SingleSource/Regression/C/uint64_to_float
-56.3165% +/- 25.4234%
SingleSource/UnitTests/Vectorizer/gcc-loops
-18.5309% +/- 6.8496%
regressions:
MultiSource/Benchmarks/ASCI_Purple/SMG2000/smg2000
18.351% +/- 12.156%
SingleSource/Benchmarks/Shootout-C++/methcall
27.3086% +/- 14.4733%
llvm-svn: 197099
For one predicate to subsume another, they must both check the same condition
register. Failure to check this prerequisite was causing miscompiles.
Fixes PR18003.
llvm-svn: 197089
